Facial expression recognition on partially occluded faces using component based ensemble stacked CNN.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Facial Expression Recognition (FER) is crucial for human-computer interaction and surveillance.
• Existing Ensemble Stacked Convolution Neural Networks (ES-CNN) struggle with occlusions, pose, and illumination variations.
• Current models often use features from the entire face, limiting performance in real-world scenarios.

Purpose of the Study:

• To propose a novel Component based Ensemble Stacked Convolution Neural Network (CES-CNN) for robust FER.
• To address the limitations of existing ES-CNN models in recognizing facial expressions under challenging conditions like occlusions.
• To enhance the accuracy and reliability of FER systems.

Main Methods:

• Developed CES-CNN, which applies ES-CNN to individual facial components (eyes, eyebrows, nose, cheek, mouth, glabella).
• Utilized subnets for each facial component to extract localized features.
• Employed a Max-Voting based ensemble classifier to combine decisions from component subnets for optimized recognition.

Main Results:

• The proposed CES-CNN demonstrated significant improvements in recognition accuracy compared to state-of-the-art models.
• Experimental validation on benchmark datasets confirmed the effectiveness of the component-based approach.
• CES-CNN showed enhanced performance, particularly on partially occluded faces.

Conclusions:

• CES-CNN offers a more robust and accurate solution for Facial Expression Recognition, especially in the presence of occlusions.
• Analyzing facial components individually leads to superior FER performance over whole-face analysis.
• The Max-Voting ensemble strategy effectively integrates information from facial components for improved emotion recognition.